PPG signal processing device and corresponding computer-implemented method

This application is a National Phase of PCT/EP2022/087548 filed on Dec. 22, 2022, which claims the benefit of priority from European Patent Application No. 21 216 999.9 filed on Dec. 22, 2021, the entirety of which are incorporated by reference.

The present invention generally relates to photoplethysmography or PPG, an optical technique to detect blood volume changes that enables to monitor various physiological parameters. The invention in particular concerns the processing of one or more PPG signals obtained for a human or animal in order to facilitate the detection of a wider variety of heart rhythm problems.

Photoplethysmography (PPG) is an optical technique that allows to monitor one or more physiological parameters by detecting blood-volume changes in the peripheral circulation. PPG makes use of light absorption by blood to track these volumetric changes. When a light source illuminates the skin, the reflected light varies as blood flows. A light sensor then converts these variations in light reflection into a digital signal, the so-called PPG signal. PPG signals are typically recorded using a pulse oximeter or photodetector, for instance the camera integrated in an electronic device like a person's smartphone, smartwatch or other smart wearable or non-wearable device.

A distinction is made between remote PPG and contact PPG. Remote PPG is non-obtrusive for the monitored person but poses major challenges to signal detection and signal processing. As a consequence, the use of remote PPG remains limited to everyday applications like leisure or fitness as its accuracy and reliability are insufficient for medical applications. Contact PPG, wherein the measurement components are in direct contact with the skin, results in a more reliable, more accurate PPG signal that facilitates medical diagnosis.

PPG can be used, among other applications, to monitor cardiovascular and hemodynamic parameters such as heart rate, heart rate variability, blood pressure, or to monitor other physiological variables such as stress, respiration or autonomic functions. One key part of an accurate monitoring with PPG is to obtain a high-quality, artefact-free signal, as PPG can be affected by various sources of noise. Methods for accurate and reliable PPG measurement are described for instance in European patent application EP3449820A1 from applicant Qompium, entitled “Computer-Implemented Method and System for Direct Photoplethysmography (PPG)”, and in European patent application EP3473173 from applicant Qompium, entitled “Computer-Implemented Method and System for Direct Photoplethysmography (PPG) with Multiple Sensors”.

In a medical context, contact PPG is mainly used for atrial fibrillation (AF) risk detection, the most common cardiac rhythm disorder. Other cardiac rhythm disorders are typically detected via an electrocardiogram (ECG) obtained for instance by a Holter monitoring system.

It is desirable to use PPG signals that are easily obtainable through wearables to detect other abnormal heart rhythms like for instance tachycardia or atrial flutter. Atrial flutter is a type of abnormal heart rhythm caused by an abnormal electrical circuit in the upper chamber of the heart, named the atria, that makes the atria beat quickly. In a normal functioning heart, electrical pulses are sent from the sinus node—the so-called SA node—in the right atrium of the heart. This way, this node controls the heart rate and timing of heartbeats. In case of atrial flutter, an abnormal electrical circuit is formed in the atria. This abnormal electrical circuit takes over the heart rhythm and rate, and causes abnormally frequent contractions in the upper chambers.

Atrial flutter also may cause the lower chambers of the heart—the so-called ventricles—to beat faster but often not as fast as the atria. In a normal functioning heart, the atrial rate or AR equals the ventricular rate or VR. In case of atrial flutter, the ventricular rate typically is an integer fraction of the atrial rate, with the ratio between the AR and VR, the so-called AV conduction rate, being equal to 2:1, 3:1 or 4:1. The AV conduction rate may be constant. In an example where the AV conduction rate remains 3:1 while the AR is 300 bpm (beats per minute), the VR will be 100 bpm. The AV conduction rate may also be alternating. In an example where the AV conduction ratio is alternating between the 2:1, 3:1 and 4:1 ratios while the AR is 300 bpm, the VR is alternating between 150 bpm, 100 bpm and 75 bpm.

On a PPG signal, the ventricular rate can be seen but the atrial rate cannot be seen. Consequently, the presence of flutter waves typically cannot be seen on a photoplethysmogram or PPG signal plot. Reliable atrial flutter detection therefore requires confirmation through an electrocardiogram or ECG.

The article “Detecting Atrial Fibrillation and Atrial Flutter in Daily Life Using Photoplethysmography Data” from the authors Eeriksinen L. M. et al., published in IEEE J Biomed Health Inform, 2020 June, 1610-1618, pretends to describe a way to detect atrial flutter using PPG. However, the features extracted from the PPG signal by Eeriksinen et al. are not particular for atrial flutter detection. Classical heart rate variability features such as entropy, RMSSD, pNN70, etc., are used, which are traditional features found in the literature. If these features would allow to reliably detect atrial flutter, this would have been confirmed in medical literature. The study underlying the article contains only 5 atrial flutter patients, 3 of which are used for training the classifier, leaving 2 patients in the test set. This gives limited insights into the performance as the limited patient size probably biases the results. Moreover, it is noticed that Eeriksinen et al. perform classification on individual 30-second strips of a PPG signal, not at the patient-level.

According to the state of the art on PPG, atrial flutter is often mentioned within a long list of arrhythmia that can be detected, but existing literature fails to describe how PPG signals must be processed to detect atrial flutter and distinguish atrial flutter from other arrhythmia.

United States patent application US2020/0100693A1 entitled “Arrhythmia Monitoring Using Photoplethysmography” describes atrial flutter” in [0070] with reference to the atrial rate and ventricular rate. In [0076] in D14 seems to teach that the average RR-interval can be used as parameter to detect tachycardias such as atrial flutter, but the document fails to teach how such RR-intervals must be processed in order to be able to detect atrial flutter.

United States patent application US2021/0007621A1 entitled “Method to Analyse Cardiac Rhythms Using Beat-To-Beat Display Plots” suggests in paragraphs [0068], [0074], [0105] and in FIG. 10 that atrial flutter can be detected from beat-to-beat display plots, but also this document contains no guidance towards preferred processing of PPG signals in order to be able to reliably detect and discriminate atrial flutter from other heart rhythm diseases.

United States patent application US2017/0032221A1 entitled “Method, Electronic Apparatus and Computer Readable Medium of Constructing Classifier for Disease Detection” refers in [0004] and FIG. 1C to a classifier for atrial flutter but relies on features of an ECG signal: disappearance of the interval between the end of the T-wave and the beginning of the P-wave.

In United States patent application US2016/0302677A1 entitled “Calibrating for Blood Pressure Using Height Difference”, paragraphs [0126], [0130] and FIG. 16B suggest to detect atrial flutter by looking for normal (or regular) heart rates and for changes in heart rate that are multiples of each other, detectable through clusters offset from the diagonal in a plot shown in FIG. 16B where RRis plot vs. RR.

United States patent application US2014/0221845A1 describes in paragraphs [0036]-[0038] obtaining a PPG signal and filtering the PPG signal to detect peaks (heartbeats) therein. US2014/0221845A1 further describes determining peak-to-peak intervals (RR-intervals). From the RR-intervals, heartrate dynamics are established, useful to detect arrhythmia. US2014/0221845A1 however does not describe a technique to detect atrial flutter.

It is an objective of the present invention to provide an improved technique to process PPG signals in order to extract information from PPG signals that is useful to detect atrial flutter or other abnormal heart rhythms which are nowadays difficult to detect or discriminate reliably based on PPG signals.

According to embodiments of the invention, the above-defined objective is realized by a photoplethysmography signal processing device as defined by claim, abbreviated as PPG signal processing device, configured to detect atrial flutter, the PPG signal processing device comprising at least one processor and at least one memory including computer program code, the at least one memory and the computer program code configured to, with the at least one processor, cause the PPG signal processing device to:

Indeed, the existence of a relationship and/or the nature of the relationship between central or averaged interbeat values for baselines (also named RR-baselines) detected based on PPG signal(s) for a single person or animal, may allow to reliably detect certain arrhythmia like for instance alternating atrial flutter. Other heart arrhythmias that could be detected this way are for instance atrioventricular nodal reentry tachycardia (AVNRT), atrioventricular reciprocating tachycardia (AVRT), atrial tachycardia, multifocal atrial tachycardia. These arrhythmias are part of the category of supraventricular tachycardia, where also atrial flutter belongs to, as explained above.

According to embodiments of the invention, one or more PPG signals are obtained for a single person or animal. In these PPG signals, heartbeats are detected using state-of-the-art algorithms that determine peaks in the PPG signals that correspond to heartbeats of the person or animal and that determine the timing of such peaks. Thereafter, interbeat intervals, also named RR-intervals or peak-to-peak intervals, are determined. This implies that for each pair of consecutive heartbeats in the PPG signal(s), the time interval between these heartbeats is determined. Instead of looking for variation in the heart rate in order to detect a heart rate variability indicative for a heart arrhythmia, the PPG processing device according to the invention then looks for interbeat interval baselines, which are short periods of time wherein the interbeat interval or heart rate remains substantially constant, i.e. constant within a predefined tolerance value for the heart rate variability in an interbeat interval baseline. Hence, all consecutive interbeat intervals that are equal within the set tolerance value together form part of a single interbeat interval baseline. According to the invention, multiple interbeat interval baselines are detected for a single person or animal, across one or plural PPG signals obtained for that person or animal. These different interbeat interval baselines shall typically have different lengths but each interbeat interval baseline must comprise a plurality of subsequent interbeat intervals that are substantially equal. It is possible that a minimum length is set for interbeat interval baselines, corresponding to a minimum amount of subsequent interbeat intervals that must be constant in order to be considered an interbeat interval baseline. Within each interbeat interval baseline, the interbeat interval is averaged, thus obtaining an average value or central value for the time in between consecutive heartbeats for each detected period wherein the heart rate remains substantially constant. Thereupon, the existence of a relationship between these averaged interbeat intervals is searched for and outputted. The averaged interbeat intervals may for instance be equal (within tolerances), the averaged interbeat values may be integer multiples of a lowest value (within tolerances), the averaged interbeat values may be integer parts of a common multiple value (within tolerances), etc.

The present invention brings the advantage that certain arrhythmias, in particular arrhythmias that are part of the category of supraventricular tachycardia, can be detected reliably from PPG signals that are collected at multiple occurrences across time for a same patient, person or animal, using convenient wearables equipped with a photodetector.

In order to establish the presence of a relationship between the central interbeat interval values determined for different interbeat interval baselines of a single person or animal, these central interbeat interval values may be compared to atrial lines. The atrial lines correspond to integer parts of a common multiple value, and the common multiple value represents an atrial rate or AR value. For a single AR value, for instance 180 bpm, the first three atrial lines correspond to 0.67 seconds (or 90 bpm, being ½ of the AR), 1.00 seconds (or 60 bpm, being ⅓ of the AR), and 1.33 seconds (or 45 bpm, being ¼ of the AR). For each central interbeat interval value, it is determined which of the atrial lines (0.67 seconds, 1.00 seconds or, 1.33 seconds) is the nearest, and the distance to the nearest atrial line is determined. This is then repeated for a subset of atrial rates, for instance 180 bpm, 182 bpm, 184 bpm, . . . , 300 bpm. A small distance between the central interbeat intervals and atrial lines of a single AR indicates that the detected interbeat interval baselines (periods wherein the interval between subsequent heartbeats or heat rate remains substantially constant) all occur when the person's atria beats at a particular rate.

In order to establish if the interbeat interval baselines detected for a single person or animal occur when the person's or animal's atria beats at a particular rate (the AR), the distances measured between the respective central interbeat intervals and the nearest atrial line for that AR are averaged. This way, a single value is obtained representing across all PPG signals the distance between the baseline interbeat intervals of a person during short periods with constant interbeat interval and the respective nearest atrial lines for a single AR. Such single value can be determined for each AR value in the subset of considered AR values.

Out of the subset of atrial rates for which the central interbeat interval values in interbeat interval baselines are compared to the atrial lines, the atrial rate is selected for which the average distance of the interbeat interval baselines to the atrial lines is the smallest. If interbeat interval baselines occur at a particular AR, the so selected atrial rate is the best candidate for the particular AR.

The PPG signal processing device may be used to detect heart arrhythmias that are part of the category of supraventricular tachycardia like atrial flutter, atrioventricular nodal reentry tachycardia (AVNRT), atrioventricular reciprocating tachycardia (AVRT), atrial tachycardia, multifocal atrial tachycardia. Embodiments of the invention may comprise a single classifier for a single arrhythmia or may comprise plural classifiers to detect plural arrhythmias. The smallest average distance between the interbeat interval during interbeat interval baselines and atrial lines for a subset of atrial rates may serve as input parameter for a classifier that determines if the person or animal suffers from atrial flutter.

A simple-to-implement atrial flutter classifier may compare the obtained smallest average distance (between the interbeat interval during interbeat interval baselines and closest atrial lines for a subset of atrial rates) to a predetermined distance threshold value. When the smallest average distance stays below this predetermined distance threshold value, an atrial rate is found for which the interbeat interval during different interbeat interval baselines approaches the atrial lines very closely. The interbeat interval baselines in other words very likely occur when the atria is beating at a particular atrial rate. The interbeat intervals in different interbeat interval baselines may differ substantially but they all closely match one of the atrial lines of that particular atrial rate. This is an indication for atrial flutter and the classifier may therefore reliably classify the person or animal as an atrial flutter patient.

It is noticed that the threshold-based classifier could be used on its own, meaning that the average distance to the closest set of atrial lines can be compared to some threshold to classify a person/animal as atrial flutter patient. If the average distance value is below the threshold, it's atrial flutter. If the average distance value is above the threshold, it's not atrial flutter. The threshold-based classifier can also be used in combination with other methods to extract information from the PPG recordings. The current method then serves to calculate a feature, the feature being the average distance to the closest set of atrial lines. This feature, along with other features computed by other methods, can then be used to train a classifier (a decision tree, a support vector machine, a k-nearest neighbour, a neural network, etc.) for one or several arrhythmias. As an example, one could determine:

In embodiments of the PPG signal processing device according to the invention, as defined by claim, the average distance corresponds to a weighted square distance wherein the length or duration of an interbeat interval baseline serves as respective weight for the square distance between the interbeat interval baseline and a nearest atrial line in the average distance.

Whereas the average distance between the central interbeat interval values of interbeat interval baselines and the respective nearest atrial lines of an atrial rate may be determined as the mean value, the median value, the modus value, etc., preferred embodiments of the invention determine the average distance as a weighted square distance value. In the weighted square distance value, each distance between the central interbeat interval value of an interbeat interval baseline and the corresponding nearest atrial line is squared and weighted with a respective weight value that is proportional to the length or duration of the interbeat interval baseline. This way, the distance value determined for a longer interbeat interval baseline gets a weight or importance in the overall average distance that is higher than the weight or importance assigned to the distance value determined for a shorter interbeat interval baseline. Weighting the distances to the nearest atrial lines using the length of interbeat interval baselines improves the reliability of the PPG signal processing device and the disease classification based thereon.

In embodiments of the PPG signal processing device according to the invention, as defined by claim, the atrial lines for an atrial rate out of the subset of atrial rates correspond to half of the atrial rate, a third of the atrial rate, and a fourth of the atrial rate.

Indeed, in case of alternating atrial flutter, it may be assumed that the interbeat intervals of RR-baselines fall on lines that are integer fractions of a base atrial rate, with the AV conduction ratio being equal to 2:1, 3:1 or 4:1. It is therefore sufficient to consider for each atrial rate in the subset the atrial lines that correspond to half, a third or a fourth of the atrial rate.

In embodiments of the PPG signal processing device according to the invention, as defined by claim, the subset of atrial rates comprises all integer rates between a lower bound and an upper bound, preferably between 180 beats per minute and 400 beats per minute.

Thus, in order to determine the atrial rate for which the interbeat intervals of RR-baselines approach the atrial lines most closely, a limited subset of atrial rates may be considered. This limited subset may contain for instance all integer values between a lowest atrial rate and a highest atrial rate. The lowest atrial rate may for instance be chosen equal to 180 bpm. The highest atrial rate may for instance be chosen equal to 400 bpm. In alternative implementations, the lowest atrial rate and/or the highest atrial rate may be chosen differently, and instead of each integer value a step value may be configured corresponding to the step or distance between two successive atrial rates considered.

In embodiments of the PPG signal processing device according to the invention, as defined by claim, the average interbeat interval value for an interbeat interval baseline corresponds to one of the following:

Embodiments of the invention determine, as explained above, an average value or central value for the interbeat interval in each RR-baseline. As the interbeat interval remains substantially constant within an RR-baseline, the interbeat values are assumed to be equal or nearly equal. Still it makes sense to determine an average or central value for the continued processing, e.g. comparison of such average or central value to atrial lines of atrial rates. The average or central value may be chosen to be less sensitive for outliers, like for instance the median value or modus value. Alternatively, since outliers may be absent as a result of the variability tolerance, the average or central value may also be determined using the mean value or the mid-range value.

Embodiments of the PPG signal processing device according to the invention, as defined by claim, further comprise means to configure a minimum amount for the plurality of consecutive interbeat intervals that forms an interbeat interval baseline.

Allowing the user to configure the minimum amount of interbeat intervals that constitutes an RR-baseline, provides the user a parameter to control the accuracy and reliability of the device. When the minimum length of RR-baselines is set higher, the accuracy will increase as the risk for considering a short period of time wherein the heart rate incidentally remains constant as an RR-baseline indicative for certain heart rhythm diseases is reduced this way.

Embodiments of the PPG signal processing device according to the invention, as defined by claim, further comprise means to configure said predefined tolerance value.

Indeed, the variability tolerance constitutes a second parameter allowing the user to control the accuracy and reliability of the device, when made configurable. A smaller variability tolerance value will result in fewer periods of time with substantially constant interbeat interval being considered as RR-baselines (due to a value outside the variability tolerance, a period of time may not reach the minimum required length to be considered an RR-baseline). Further, within RR-baselines, the interbeat interval values will in general be closer to each other (they can differ no more than the variability tolerance). The risk for considering a short period of time wherein the heart rate incidentally remains constant as an RR-baseline indicative for certain heart rhythm diseases is thus reduced when the variability tolerance is decreased, or vice-versa.

In addition to a PPG signal processing device, the present invention also concerns, as defined by claim, a corresponding computer-implemented method for processing a photoplethysmography signal, abbreviated as PPG signal, to detect atrial flutter, the method comprising:

According to a further aspect, the present invention also concerns a computer program product as defined by claim, comprising computer-executable instructions for performing the method according to the invention when the program is run on a computer.

According to yet another aspect, the present invention concerns a computer readable storage medium as defined by claim, comprising the computer program product according to the invention.

illustrate an embodiment of the PPG processing device that is configured to detect RR-baselines and compare such RR-baselines with atrial lines corresponding to atrial rates in order to detect atrial flutter.

In a first step, denoted, the PPG processing device obtains one or plural PPG signals for a person.shows a PPG signalspanning 60 seconds that is obtained for a person in step.

In a second step, denoted, the PPG processing device detects heartbeats in the PPG signal. In, the dots,,,,, etc., represent heartbeats that are detected in the PPG signal. These heartbeats are detected through state-of-the-art peak detection algorithms applied to PPG signalor to portions of PPG signal.

In a third step, denoted by, the PPG processing device determines the length of interbeat intervals for pairs of consecutive heartbeats in the PPG signal. These interbeat intervals may be outputted or plotted in a tachogram like the one shown in. In, dotrepresents the interbeat interval between the first heartbeatand the second heartbeatdetected in the PPG signal. The interbeat interval corresponds to the time lapsed in between the heartbeatsand, and is expressed in seconds. From, it is seen that the interbeat intervalbetween heartbeatand heartbeatcorresponds to 0.7 seconds. Similarly, the trianglerepresents the interbeat interval between the second heartbeatand the third heartbeatdetected in PPG signal. This second interbeat intervalcorresponds to 1.40 seconds. Further, trianglerepresents the interbeat interval between the third heartbeatand the fourth heartbeatdetected in PPG signal. This third interbeat intervalcorresponds to 1.45 seconds. Triangleinrepresents the interbeat interval between the fourth heartbeatand the fifth heartbeatdetected in PPG signal. This fourth interbeat intervalcorresponds to 1.42 seconds. Each subsequent dot or triangle incorresponds to an interbeat interval between a pair of consecutive heartbeats detected in the PPG signal. When plotted on a timeline, as is done in, these interbeat intervals constitute a tachogram.

In a fourth step, denotedin, the PPG processing device determines RR-baselines. An RR-baseline corresponds to a plurality of consecutive interbeat intervals that are equal (within a variability tolerance). In, the detected RR-baselines are represented by triangles. Interbeat intervals that are not part of an RR-baseline are represented by dots. The interbeat intervals,andfor example constitute a first RR-baseline, because they are consecutive interbeat intervals and their respective values vary less than 0.05 seconds (which is assumed to be the variability tolerance in). In addition to the RR-baseline formed by the interbeat intervals,and,shows five additional RR-baselines detected based on PPG signal. Two parameters define the accuracy of the RR-baseline detection step: the minimum length of an RR-baseline (set at 3 interbeat intervals in) and the variability allowed within an RR-baseline (set at 0.05 seconds in). Increasing the minimum length of an RR-baseline and/or decreasing the variability tolerance shall improve the accuracy and reliability of the PPG processing device as these parameter modifications will result in less time intervals wherein the heart rate accidentally remains constant being considered as RR-baselines (less false positives).

In a fifth step, denotedin, the PPG processing device determines a central interbeat interval value for each RR-baseline detected in step. The central value or average value may for instance correspond to the mean value of all interbeat interval values of the RR-baseline. Alternatively, the median value, the modus value, the midrange value, or another central value may be considered. In the example of, the mean value is determined as the central value for the interbeat values of an RR-baseline. For the RR-baseline that contains the interbeat intervals,,, this central value hence corresponds to (1.40+1.45+1.42)/3=1.42 seconds. Similarly, the central baseline value is determined for each of the detected RR-baselines. These central baseline values, obtained for a single person across one or plural PPG signals, may then be plotted against an index, as is done inand.

Depending on the arrhythmia or arrhythmias looked after, the central RR-baseline values are then analysed in order to establish if certain relationships exist. This is indicated byin. In an example where atrial flutter is looked after, the PPG processing device may for instance be configured to detect if the central RR-baseline values substantially correspond to alternating integer parts (atrial lines) of a common multiple value (atrial rate), where the alternating integer parts either represent half, a third or a fourth of the common multiple value. The existence of other relationships between the central RR-baseline values may be investigated if other arrhythmias are to be detected by the PPG signal processing device.

In order to compare the RR-baselines with atrial lines the distance between the central interbeat value of a detected RR-baseline and the respective atrial lines corresponding to an atrial rate is determined, and the nearest distance is maintained. This is illustrated bywhereinrepresents the central interbeat value of a first RR-baseline,represents the central interbeat value of a second RR-baseline, etc. In, the atrial rate of 300 bpm is considered and the three atrial lines corresponding to this rate are represented by dashed line(half of the atrial rate corresponds to 150 bpm or 0.40 seconds interbeat interval), dotted line(a third of the atrial rate corresponds to 100 bpm or 0.60 seconds interbeat interval), and dash-dotted line(a fourth of the atrial rate corresponds to 75 bpm or 0.80 seconds interbeat interval). For the first RR-baseline, the nearest atrial line to central interbeat interval valueis line, so the distance between central interbeat interval valueand atrial lineis maintained. For the second RR-baseline, the nearest atrial line to central interbeat interval valueis line, so the distance between central interbeat interval valueand atrial lineis maintained. Similarly, the distance between each central interbeat interval value, represented by a dot in, and the atrial lines,,is determined, and the smallest distance, i.e. the distance to the nearest atrial line, is maintained. In stepof, this process, which is illustrated infor the atrial rate of 300 bpm, is repeated for a subset of atrial rates, for example all integer atrial rate values between 180 bpm and 400 bpm.